Excerpt
Contents
List of Figures
List of Tables
1 Introduction
2 An (inter-)national Perspective
3 Data Description
3.1 The Data Set
3.2 Variable Specification
4 Empirical Analysis
4.1 Changes in the Workforce Composition
4.2 Evolution of Weekly Working Hours
4.3 Trends in Part-Time Employment
4.4 Hours, Wages, and the Inequality of Incomes
4.5 Discussion
5 Summary and Conclusion
A Appendix
A.1 OECD Statistics - Average Annual Hours
A.2 Construction of Hours Measure
A.3 Tables and Figures
Bibliography
List of Figures
1 Kernel Density Estimations for Average Weekly Working Hours
2 Average Weekly Working Hours for Men and Women
3 Percentage Changes of Working Hours at Different Percentiles of the Distribution: 1984 - 1996 and 1997 - 2009
4 Cross-Sectional Age Profiles of Average Working Hours
5 Average Working Hours across Different Skill Groups
6 Share of Part-Time Employment in Total Workforce and Work Volume
7 Evolution of Average Wages and Employment Shares of Part-Time and Full-Time Workers
8 Indexed Growth of Incomes, Wages, and Hours at Different Percentiles
9 Variance Decomposition of log Monthly Incomes
10 Variance Decomposition of log Monthly Incomes: A Comparison to Full-Time Employed Workers
A.1 Percentage Changes of Incomes and Wages at Different Percentiles of the Distri- bution: 1984 - 1989, 1990 - 1999, and 2000 - 2009
A.2 Gini of Monthly Incomes versus Gini of Hourly Wages
List of Tables
1 Composition of Overall Labour Supply
2 Percentage Shares of Part-Time Employed Workers within Age and Skill Groups
3 Composition of Part-Time Labour Supply
4 Example for Overstated Wage Inequality
A.1 Average Annual Hours of 13 OECD-Countries
A.2 Comparison of Different Hours Measures for Men
A.3 Comparison of Different Hours Measures for Women
1 Introduction
On Saturdays, Dad ’ s mine.
(Confederation of German Trade Unions promoting the 5-day work week, 1956.)
Looking back, this slogan may be considered the starting point for a prolonged reduction ofworking hours in the Federal Republic of Germany (Germany, henceforth) and the great majorityof industrialised countries, which has lasted through the second half of the twentieth century.
Because of the breadth of the topic and its significance for economic welfare, the realmof working hours has set the agenda for sustained public, political, and scholarly debates ona national and international scale. Not only are the hours of work a principal component toproductivity, they also represent a key ingredient to an individual’s quality of life and a society’swell-being. In spite of this, thorough documentations of recent changes in the distributionof hours are relatively scarce, in particular, if compared to the abundance of research that theeconomic profession has directed towards the dissection of the structure of wages. Using Germansurvey data for the period from 1984 to 2009, my thesis therefore sets out to do three things:firstly, to document recent changes in the distribution of hours; secondly, to describe the relatedtrends in part-time employment; and thirdly, to elaborate on the interrelation between hours,wages, and incomes in determining economic inequality.
Broadly speaking, the distribution of hours in industrialised countries has become morediversified. Specifically, the range of hours that we observe, and the relative frequency at whichthey occur, have both changed considerably over time. These changes give rise to sizeableramifications for economic welfare and inequality. In particular, they imply that a progressivelydecreasing share of the labour force works within the framework of a standard employmentrelationship - Normalarbeitsverh ä ltnis - characterised by indefinite, dependent, full-time work.The concurrent surge of non-standard, or atypical, forms of employment necessitates a steadyadjustment of the legal environment, and hence requires a detailed comprehension on the partof economists and policymakers alike.
Changes in the distribution of hours come from many sources, but can be organised alongtwo broad dimensions: changes in the workforce composition - that is, adjustments in theobserved demographic structure of the workforce - and changes in the idiosyncratic distributionof hours within each demographic group.[1] During recent decades, the German labour markethas been marked by substantive changes along both of these dimensions. In terms of events anddevelopments that impinged upon the workforce composition, Germany has experienced, amongother things, a prominent rise in female employment rates, an ageing population, an increase inthe supply of high-skilled workers, German reunification, and several waves of immigrants.[2]
At the same time, the working hours within each of the existing, and likewise within each ofthe emerging groups started to diversify. This diversification received additional impetus fromthe changing objectives of trade unions: whilst they fought for a standardisation of workinghours until the 1980s, they promoted an individualisation and decentralisation of hours in theyears that followed (cf. Messenger (2007)). In other words, they began to advocate a work-lifebalance based on individual as opposed to collective preferences, and thereby enabled alreadyemployed persons as well as hitherto untapped parts of the labour force to realise, or at leastcome closer to, their desired individual trade-off between working time and leisure time. As amajor consequence of this gain in flexibility, female employment rates surged, and dual earnerhouseholds became the rule rather than the exception.[3] However, these unequivocal benefitsentailed a number of costs which materialised in, for instance, a stronger gender segregation andmore fragile employment relationships - the latter often reflected in shorter job tenure. Whereaswomen had previously been rarely present in the labour market, but if so then mostly workingfull-time, they now began flowing into a variety of new professions, albeit chiefly under part-timearrangements. Men, in contrast, who had been working full-time all the while, continued to doso. In addition to this, part-time workers in general started to face notable disadvantages in thelabour market, for example, by way of lower hourly wage rates, depressed career opportunities,lower pensions, and reduced job-related training.[4]
An ageing society, in conjunction with persistent technological progress and increasing global competition ushered in a growing scarcity of labour in the past, and thereby reinforced the demand for rising participation rates of women, and other, currently unused reserves of labour resources. The same forces will almost inevitably require a further expansion of jobs outside the framework of the classical full-time work relationship, and thus entail a sustained diversifiation of working hours for years to come.[5]
The above considerations imply critical changes in economic welfare, and thereby demon-strate that it is crucial to carefully understand the consequences of these changes for the variousfacets of economic inequality. They concurrently propose the German labour market as a promis-ing object of study. Together, this poses the starting point for my thesis in which I strive toexplore empirically three core aspects of economic inequality: inequality of hours, inequalityof wages, and inequality of incomes. Building on microdata from the German Socio-EconomicPanel (henceforth, GSOEP) for the period between 1984 and 2009, I aim to address four closelyrelated questions. First, what are the general trends in the distribution of hours, and how dochanges between and within demographic groups contribute to them? Second, what are thecharacteristic features of part-time employment, and have they changed over time? Third, howdid the distribution of wages and incomes change, and what is the link to trends in the distri-bution of hours? And fourth, can an analysis of trends in income inequality reasonably neglectchanges in the inequality of hours?
The last question connects to a recurring discourse about potential mismeasurement of trends in income inequality that might occur when using the Sample of Integrated Labour MarketBiographies (SIAB, henceforth) provided by the Research Data Centre of the German FederalEmployment Agency at the Institute for Employment Research. To put this into some context,the GSOEP data set is an annual survey comprising self-reported information on monthly wagesand weekly working hours, and thus readily permits the computing hourly wage rates.[6] Incontrast, the SIAB data contain daily wages, but no information on the corresponding workinghours, except for a classification into part-time and full-time employment. Because daily wagesstand in a fixed relation to monthly incomes, disregarding hours might lead to a biased estimationof trends in income inequality. To see this point more clearly, one may conceptualise monthlyincome more generally as the result of multiplying monthly hours of work (quantity) by thehourly wage rate (price).[7] Now, if the observed income is fixed because the reported dailywage is the same, but the person is working more (less) hours, the implicit hourly wage ratemust decline (increase). This implies an overestimation (underestimation) of income inequality,compared to inequality of hourly wages when the worker belongs to the top of the incomedistribution. In order to obtain an unbiased estimation of economic inequality measured byinequality of incomes, it is therefore essential to observe changes in both the structure of wages and the structure of hours, unless the distribution of hours for the investigated group does not change.[8] The latter, qualifying assumption is usually presupposed in SIAB-based analyses, butit seems that a broader empirical validation, at least for the most recent years, is still missing.My thesis endeavours to inquire whether it is worthwhile to expend more research effort in thisdirection.
The road map for my empirical exploration is inspired by the two dimensions I outlined earlieron: changes in the relative size between distinct groups of the labour market, and changes inthe distribution of hours within each individual group. After gaining a thorough comprehensionof these trends, and how they jointly shape the dynamics of the overall supply of hours, I willdevote special attention to the role of part-time employment as a working arrangement that hasbecome widely accepted and administered in recent decades. In doing so, my aim is not only tounderstand the role of part-time work for the aggregate hours supply, but also to discern whoworks in part-time and how this composition may have changed through time. Informed by theresults, I will then proceed to explore the linkage between inequality of hours and inequalityof wages, and examine how they jointly determine the distribution of incomes. Finally, I willuse my findings to delineate and discuss the concern of potential mismeasurement of incomeinequality when grounding an analysis in the SIAB administrative records.
My results indicate a broad and prolonged decline of average weekly working hours between1984 and 2009, yet this degression affects men and women, different ages, and distinct educationgroups to varying degrees. Average working hours show a conspicuous reduction from around 39 . 6 to 37 . 1 between 1984 and the early to mid-2000s. The emergence of distinct trends betweenmen and women throughout the 1990s entails a rise in the gender gap of average working time by approximately 1 . 5 hours per week. Among both sexes, it is the youngest and least educated whoexhibit the sharpest deterioration, first, in terms of their absolute and relative workforce repre-sentation, and second, in terms of their average supply of working hours. Part-time employmentshows a salient rise among both genders between 1984 and 2009. In particular, more than 50percent of the entire workforce expansion of around 4 . 49 million workers manifests in part-timejobs. During the 2000s, more than 40 percent of all employed women hold a part-time position,and roughly 90 percent of all employees in part-time arrangements are female. Part-time jobshave attracted young and low-skilled workers of both genders, and, as a consequence, the ageingand upskilling, which is observed for the workforce as a whole, is notably less striking for thesubset of part-timers. The average wage of part-time workers of either gender exhibits a con-tinuous decline throughout the period under consideration, and levels around 87-88 percent ofthat of full-time workers during the 2000s. Among men, income inequality is almost exclusivelyshaped by trends in the dispersion of wages. As a result, both the income and the wage distri-bution of men reveal the same trends; in particular, they both show a rapid growth in inequalitystarting during the mid-1990s. Among women, by contrast, changes in hours and wages alikedetermine the distribution of incomes. The female wage structure is compressed until the year2000, and the ensuing dispersion is led by lower-tail inequality. The female income distribution,by contrast, features two periods of growing inequality: the first ranges from 1992 to 2000 andis dominated by the lower-tail, and the second lasts from 2000 to 2009 and is effected by boththe lower- and the upper-tail. Altogether, the results indicate that an analysis of trends inearnings inequality can safely abstract from changes in the variation of hours as long as the unitof analysis is a West German full-time worker - notably of either gender - within the age rangeof 21 to 60 years.
My thesis unfolds as follows: Section 2 provides a synopsis of trends in aggregate laboursupply and establishes a few indicative facts for the distribution of hours in Germany. Section 3introduces the data set and specificies the sample as well as the variables used for the analysis.Special attention is devoted to the construction of the hours variable as it is vital to the successof my examination. The empirical analysis is presented in Section 4 and structured in fiveparts: Sections 4.1 and 4.2 commence by presenting trends in the workforce composition andthe distribution of working hours within different demographic groups. Section 4.3 then turnsto the role of part-time employment. This is followed by an exploration of the link betweenhours, wages, and incomes in Section 4.4, before Section 4.5 completes the empirical part byrelating the results to potential measurement error in the SIAB data. Section 5 summarises and concludes.
2 An (inter-)national Perspective
To set the scene and to embed the following microeconomic analysis, it is useful to commencewith a brief synopsis of recent trends in aggregate labour supply, and to inquire prima facie evidence of concurrent changes in the distribution of working hours in the present microdata.
The conspicuous decline of average working hours is a phenomenon that has been observedfor industrialised countries all around the globe. In particular, comparisons of average annualworking hours based on OECD statistics for the years between 1980 and 2011 establish the following key facts.[9] Starting with Germany and France, the data show an average annualnumber of working hours of 1,751 and 1,795 for the year 1980. By 2011, these figures are downeach by roughly one-fifth, to 1,399 and 1,476 hours, respectively. Quite similar trends, albeitcommencing from a lower level, can be discerned for the Netherlands: down by roughly 11percent, from 1,553 to 1,379 average hours per year. And the list goes on, including Italy (from1,859 to 1,774), Great Britain (from 1,767 to 1,625), Denmark (from 1,659 to 1,522), Norway(from 1,580 to 1,426), Japan (from 2,121 to 1,728), and so forth, except for two countries, theUnited States (US, henceforth) and Sweden. With figures for the US revolving between 1,767and 1,849 hours over the whole period, it is, in fact, Sweden that can be singled out as the onlycountry in which annual hours display a sustained expansion: between 1980 and 2011, averageannual working time increases by 10 percent, from 1,517 hours to 1,644 hours. Finally, it iscompelling to juxtapose these figures with data for the Republic of Korea: during the 1980s,a Korean worker worked between 2,850 and 2,900 annual hours, hence almost twice (!) thenumber of a Swedish worker at that time. In 2011, the Korean figures are still well above 2,000hours, yet around 28 percent below their peak in 1986 (2,911 hours).
What explains the long-standing reduction of working hours? Finding an answer to this question requires a deeper exploration of the forces which determine the shape and dynamics of the distribution of hours. In other words, one must investigate the labour supply decision which is taken by individual economic agents. As a first step to this end, I shall divert my attention away from average annual hours to average weekly hours.[10]
The individualisation and flexibilisation of working time arrangements, abetted by the intro-duction of flex-time and working time accounts, essentially provoked both a persistent reductionof average hours, on the one hand, and a simultaneous diversification, or polarisation, of workinghours on the other hand. The impact of these changes, especially of the growing diversity ofworking hours, is exemplified in Figure 1 which illustrates the distribution of average weeklyworking hours in the German labour market during three distinct periods: 1984-1987, 1995-1998, and 2006-2009.[11] To begin with, the male hours distribution disperses considerably, butthis dispersion is more pronounced during the earlier period, ranging from the mid-1980s untilthe mid to late 1990s. As opposed to this, the later period, itself lasting until the late 2000s, ismarked less by an overall diversification of working hours among men than it is characterised bythe emergence of several small accumulations in the right tail of the density. Given the temporalpatterns, one may hypothesise that the initial dispersion partially reflects the gradual disman-tling of the standard 40-hour week by collective agreements that accomplished hours reductionsdown to 35-hour weeks during the early to mid-1990s. Probably among the most prominentof these reductions is the ‘35-hour week’ enacted in the metal industry in 1995. As the his-tory of collective agreements in Germany during those years reveals, however, this particular
Abbildung in dieser Leseprobe nicht enthalten
Figure 1: Kernel Density Estimations for Average Weekly Working Hours
Notes: The three plots depict weighted kernel density estimations for the distribution of average weekly workinghours aggregated over 4-year intervals. Each panel shows the the individual densities for men and women, aswell as for the pooled sample. The sample contains all persons aged between 21 and 60 who work between 5and 85 hours per week, and report positive earnings. All graph axes are equally scaled, and all computations usecross-sectional sample weights.
Source: Own computations based on GSOEP data for the years 1984 to 2010.
achievement for workers in the metal industry is merely the pinnacle of prolonged bargainingprocesses accentuated by numerous preceding working time concessions made by firms in a va-riety of industries (cf. WSI Tarifarchiv / Hans Boeckler Stiftung (1994-2012)).[12] This said, onecan next observe that the pooled hours distribution increasingly adapts to an equally weightedaverage of the depicted male and female hours densities. Initially carefully tracing out the hoursdistribution of men, the pooled density features distinct characteristics of both the male andthe female hours distribution by the late 2000s. This probably reflects a rising participation ofwomen in the labour market. Finally, it is worthwhile to note the bimodal shape of the femalehours distribution that becomes globally less peaked over time. This last finding illustrates theinnately larger, and further growing dispersion of hours among women as compared to men.Taken together, the observed patterns describe a steady dispersion of the hours distribution inthe German labour market as a whole. The implied growth in hours inequality raises furtherquestions concerning the origins, and, of course, the linkage to other concepts of economic in-equality, most importantly, wages and incomes. To explore these and related issues in greaterdetail is going to be the objective of the remainder of this thesis.
3 Data Description
3.1 The Data Set
My analysis is based on the full public use file of the GSOEP and covers the period from1984 to 2009. The GSOEP is an annual survey which collects information from individualsand households with the aim of providing researchers with detailed microdata, embracing the entire spectrum from demographic and labour market characteristics, to financial backgroundand health status, to personal satisfaction and indicators for the shape of individual preferences.The survey secures basic coverage of these topics by way of asking a repeated set of questions ineach year. This recurring set is supplemented by detailed questions on a wave-specific, topicalmodule. Together, this design enables researchers of various professions to draw on both a set ofbaseline variables to conduct longitudinal analyses and a number of one-time-only, or irregularlyqueried, variables for the primary purpose of cross-sectional analyses (cf. Haisken-DeNew andFrick (2005)).
The GSOEP project was initiated in Germany in 1984, covering roughly 12,000 individualsin 6,000 households, and later, in the spring of 1990, extended to include the region of EastGermany. In order to sustain representativeness of the sample and to further increase thesample size, two major refreshments were performed in the years 1998 and 2000. As a result,the most recent survey waves contain more than 20,000 individual respondents in nearly 11,000households.
An instrumental and likewise crucial feature of the GSOEP survey strategy is to system-atically oversample particular subgroups of the population, such as East Germans, top incomeearners, or selected groups of immigrants. The aim of this exercise is to render empirical explo-rations of relatively small subgroups of the population (statistically) meaningful by warrantinga sufficiently large number of observations. Yet, it implies that the plain sample compositionfails to represent the actual structure of the population and, therefore, the workforce which rep-resents the object of my study. To recover the underlying workforce composition, the GSOEPprovides cross-sectional survey weights. The principal idea of these weights is straightforward:they reflect the hypothetical number of individuals, or households, that a single correspondingsurvey participant represents. Throughout the present analysis, I consistently apply these cross-sectional sample weights. Besides delivering more representative results, they further enable meto obtain comparative numbers for the absolute size of the workforce, and thereby expedite theprocess of data validation.
3.2 Variable Specification
The unit of analysis in the current study is an employed individual of working age with positive earnings. In turn, I will delineate how these three features are defined, and describe howthe corresponding variables are constructed. The unprocessed sample contains data for WestGermany between 1984 and 2009. From the GSOEP database, I retrieve unbalanced, individual-level data for both genders in private and non-private households. Importantly, this includesimmigrants as well as East Germans if their residential area at the time of survey is WestGermany.
Hours
An accurate measurement of working hours is pivotal to the success of my analysis. Yet, theavailability of hours data is generally poor, and their validation, when information is available,is involved (cf. Bound, Brown, and Mathiowetz (2001)). The GSOEP data bear the greatadvantage of providing such data on working hours on an individual level. However, being a survey study, the risk of measurement error is inherently high, and, as several US-based studies suggest, self-reported hours tend to overstate actual working hours, irrespective of the precise phrasing of the question, or the time horizon requested. Remarkably, Bound et al. (2001) review that, out of ten then available studies which exploit either household time diaries or employer records as a source of validation, ten (!) find evidence for an overestimation of self-reported hours obtained from household surveys. To better understand the possible perils exposed by the present data set, and to arrive at potential means for improvement, it seems useful to assess the issue of gauging working hours in the present data set more closely.
The GSOEP data readily provide two principal concepts of weekly working hours, namely, average actual (or effective) hours, and standard (or contracted) hours, plus a separate variable containing the weekly overtime.[13] Because I aim to probe the distribution of working hours in the way they are supplied rather than how they should be supplied, the inclusion of overtime constitutes a minimum requirement for a reasonable hours measure in the current analysis. The available variables offer two principal possibilities to meet this requirement.[14]
The first option is to rely on the self-reporting of the survey respondent with regards to thenumber of hours worked during an average week. This value is reported in the variable average actual hours. To assess potential measurement issues, I start by examining the relevant surveyquestion, which reads: ” And how much on average does your actual working week amount to, with possible overtime? ” (Question 49, Survey Wave 1995). Unfortunately, the characteristicsof this question give space to two crucial sources of measurement error. First, the question lacksinstructions, or guidance, for the respondent on how to estimate the average actual workinghours. Second, the question fails to detail a specific point or period in time to refer to as a basisfor the response. Since workers presumably abstract from factors like sick time, vacation, orovertime compensation through time off when thinking of an average work week, this questionis unlikely to elicit an accurate measurement, but rather an overstatement of working hours.If, for instance, the survey had instead asked the respondent to report the number of hoursworked during the last week - thus defining a specific reference period - then much of theambiguity would be cleared away. To see this point more clearly, notice that employees whohad a day off, or worked fewer hours during that period, say, as a compensation for previouslyaccumulated overtime, will, on average, balance with workers who had worked extra hours, andthus accumulated, or ‘banked’, overtime on so-called working time accounts.[15]
As an aside, it is worthwhile to acknowledge the issue of commuting, or, more generally, thequestion of when and where working time is defined to begin and end. While one worker mayperceive the time spent on commuting to and from work as factual working time, another mightnot, even if neither of them receives an explicit compensation for incurring the cost of forgone leisure time. As one may reasonably suspect those workers for whom commuting is officially part of the working time to acknowledge this in their estimate, this effect is likely to aggravate the overestimation of working hours in the sample.[16]
An alternative way to obtain an estimate of hours that embodies overtime is to compute thesum of standard hours and weekly overtime. This strategy capitalises on two factors which Iwould expect to improve the accuracy of the resulting hours measure. To illustrate the reasoning,let me again return to the respective survey questions. The question on standard hours reads asfollows: ” How many hours per week is your agreed working week without overtime? ” (Question48, Survey Wave 1995). And the question on overtime goes: ” What was it like last month: Did you do any overtime then, and if so how much? ” (Question 51, Survey Wave 1995). Thefirst question requests the agreed upon working time, a value which is regularly stipulated inan employment contract, and therefore probably known with reasonably high precision by themajority of workers. The subsequent question on overtime defines a specific time frame, andthus benefits, in theory, from the outlined countervailing effects of longer and shorter hoursreported by different employees that should, on average, offset each other. In practical supportof this, Hunt (1999) asserts that the overtime hours obtained from this question line up well(in total) with the overtime documented in SIAB data, a source which is widely considered todeliver fairly accurate validation data for labour market outcomes.[17] By way of contrast, ifovertime is derived from the difference in average actual and standard hours, so Hunt (1999)puts forward, the resulting figures are perceptibly larger, suggesting that self-reported averageactual hours tend to exacerbate the overstatement of working hours.
With these considerations in mind, it is interesting to examine what the data at hand suggest.The entries of columns 2 to 7 and 8 to 13 of Appendix Table A.2 (Men) and A.3 (Women)display annual summary statistics for the aforementioned hours concepts, average actual hours,and standard hours plus overtime, respectively. The range of hours for both genders is boundfrom below at 5, and from above at 85 hours per week.[18] A comparison of the first momentsamong men affirms a positive and persistent difference between the average actual hours and themeasure of standard hours plus overtime. More precisely, average actual hours exceed standardhours plus overtime by roughly 1 to 2 hours. For women, the entries show the same pattern,although the difference in means is very close to zero. One should note that the questionon overtime was not asked regularly between 1984 and 1987. Instead, in 1984 and 1985, theovertime variable was constructed indirectly as the difference between the average actual hours and standard hours (cf. DIW Berlin (2012)). This explains why the two measures are almost identical in those two years. 1986 is the first year when the questionnaire explicitly requested the survey participants to report their overtime hours, however, there is no information provided for 1987.[19] The latter is clearly visible in a sudden drop to below 40 hours among men. From 1988 onwards, the overtime variable is consistently available.
Together, the theoretical considerations and the data appear to imply that the hours mea-sure based on standard hours delivers a more accurate depiction, and thus a more conclusiverepresentation of the true hours distribution. However, relying on this measure alone impliesa conspicuous decline in the annual sample size: as a comparison of the numbers in AppendixTable A.2 and A.3 reveals, this loss is particularly strong among men, ranging from 2 . 8 percentin 1987 to more than 28 . 2 percent in 2001. In an attempt to strike a balance between the ac-curacy of the hours data and the loss of observations, I combine the two hours concepts into athird one, giving priority to the standard hours based measure. The algorithm used to computethe new hours construct is detailed in Appendix A.2, and the corresponding summary statisticsare tabulated in columns 14 to 19 of Appendix Table A.2 and A.3, respectively. In short, theconstruction starts from the standard hours plus overtime measure, and successively performs aseries of replacements. Most notably, whenever the data constellation is such that a survey par-ticipant reports zero or missing overtime, but average actual hours that exceed standard hours, Idecide to substitute average actual hours for the standard hours measure. This combined hoursvariable forms the foundation for the following analysis.
Income and Wages
The analysis uses current gross monthly labour income as a baseline variable for individualearnings.[20] To improve the representativeness of this measure, I additionally retrieve the full setof bonus payment variables from the database and add one-twelfth of the resulting aggregateannual bonus payment to the gross monthly labour income.[21] In doing so, I presuppose uniformlydistributed bonus payments over the year. The inclusion of bonus payments is for two reasons.First, workers, especially if salaried, often think in terms of annual earnings, and consequentlyincorporate annual bonus payments into their individual labour-leisure trade-off at each point intime. And second, the consideration of bonus payments helps to sidestep potential adjustmentsin the compensation structure that might occur between different points in time; for instance,away from regular income towards more performance-oriented pay schemes which, in general,draw more heavily on bonus payments.
To derive the hourly wage rates from monthly income, I use the constructed variable of weeklyhours in combination with the assumption of 4.33 weeks per month. By construction, this pin- points two important sources of measurement error. First, if working hours are confounded withmeasurement error, then so will be the corresponding wages. As mentioned before, individualreportings on working hours are probably overstated. This upward bias in hours will lead toa downward bias in wages, conditional on income. A second source of measurement error isexposed by the income variable itself, however, to sign the direction of this error is arguablymore difficult than in the case of working hours. Using administrative data as a source of vali-dation, some US-based studies find a moderate overstatement, others a modest understatementof monthly income obtained from household surveys (cf. Bound et al. (2001)). Taken at facevalue, this appears to imply that the persistent upward bias in hours overrides measurementerror in incomes, and hence wages will, by tendency, suffer from downward bias.[22]
To warrant incomes and wages comparable across years, I deflate all nominal values using the Consumer Price Index (CPI) provided by the Federal Statistical Office for the period from 1984 to 2009 (Index: 2005 = 100). Finally, I carefully trim the gender-year specific income and wage distributions below the 1st and above the 99th percentile.
Demographics
My sample is stratified by gender, eight age groups, and three skill groups. These basic demo-graphic clusters form the primitives along which the following empirical analysis is organised.Each age group covers a five-year interval, and the range of ages under consideration reachesfrom 21 to 60 years. Workers are further classified according to their highest educational degreeinto one out of three skill groups, denoted as low-, medium-, and high-skilled. The variable uponwhich I perform the classification uses the International Standard Classification of Educationscheme of 1997. Building on this variable, I define the skill groups as follows: school drop-outsand employees with general elementary schooling are classified as low-skilled. Workers withmiddle vocational degrees and vocational degrees in conjunction with Abitur are allotted to themedium-skilled. Workers that hold either a foremen education, an accomplished civil servicetraining, credentials from a polytech, or a university degree are defined as high-skilled.[23] Thelatter sort of classification is appropriate in the German context as the necessary qualificationfor a large number of jobs, primarily technical ones but also in the public service, is obtainedfrom vocational training in Germany, instead of from colleges or universities as in most othercountries (cf. Fuchs-Schündeln et al. (2010)). An important labour market group which I intentionally exclude from the sample, irre-spective of reported working hours, are the self-employed. Compared to workers in dependent employment, the working patterns of self-employed expose substantial differences, in particular,along the dimensions of the number and dispersion of working hours: both are notably larger(cf. European Commission (2010), and Lee, McCann, and Messenger (2007)). The literatureproposes two possible explanations for this finding. First, self-employed workers have a greaterleeway in displacing working hours, an opportunity that, in turn, should decrease the disutilityof work, and consequently increase the optimal amount of time spent on it.[24] And second,working longer hours may serve as a means to recompense a self-employed worker for bearinga higher job-insecurity (cf. Lee et al. (2007)). Additional hours of work would then be con-ceived of as an insurance against, for example, slipping into the state of unemployment. In lightof this, an exploration of both working hours and working patterns of self-employed personsseems a worthwhile undertaking in its own right. However, since the underlying mechanismswhich determine, or at least affect, the working time decision of self-employed appear to differsubstantially from those that influence the decision of labour market participants in dependentemployment, I exclude them from the analysis.
4 Empirical Analysis
The main purpose of this section is to foster an understanding of cross-sectional trends in the distribution of hours, wages, and incomes observed for the German labour market over the past twenty-five years. Changes in the overall distribution of hours are the result of adjustments in the demographic structure of the workforce, and changes in the hours distribution within each demographic group. These two dimensions guide the first part of the analysis.
The plan is as follows. First, I will present the main demographic characteristics that de-scribe the compositional dynamics of the sample over time. Next, I will examine trends in thedistribution of hours within these demographic groups, and offer some preliminary interpreta-tions of the findings. The role of part-time employment will lead the discourse, and supportiveempirical facts will be established. Informed by the results, I will then proceed to the dynamicsin the wage and income structure, and outline, how the detected trends in hours relate to them.The analysis of this issue is complemented by a decomposition of the variance of log incomes intoits three building blocks, the variance of log wages, the variance of log hours, and the covariancebetween hours and wages. The section closes by relating the empirical findings to the discussionof potential bias caused by insufficient control over hours in the SIAB administrative records.
4.1 Changes in the Workforce Composition
To set the stage, Table 1 reports trends in the workforce composition of men and women, alongwith the pooled sample over time. The table entries reflect three-year averages, where themidpoint labels the columns. Exceptions to the rule are 1984 and 2009, for which the averageis computed only over two years.[25] The tabulated figures mirror the composition of the activecivilian workforce subject to the sample constraints as described in the previous section.
The table establishes several key facts. To begin with, the overall number of employedworkers increases substantially throughout the period under consideration: between 1984 and2009, total employment raises by roughly 4 . 49 million workers, or 22 . 6 percent, from 19 . 91 to 24 . 40 million. A closer look at the corresponding employment figures for men and women revealsthat this growth is dominated by an overwhelming expansion of women in the labour market.More precisely, female employment figures grow by 4 . 07 million, or 52 . 4 percent, from 7 . 76 millionto 11 . 83 million. In the meantime, the number of male workers grows from an initial 12 . 15 millionvia 13 . 41 million in 1994 to 12 . 57 million in 2009. The surge of female employment is likewisereflected in an overall increase in the share of women in total employment by 9 . 5 percentagepoints. Specifically, with 48 . 5 percent representation in the labor force, women are nowadaysalmost at par with men.[26] As opposed to this, and concurrently posing a first indication forthe difference in individual labour supply between men and women, the share of total hours(work volume) supplied by women falls short of the surge in their workforce representation: in1984, women account for roughly one-third of the total hours supply in the labour market, whileamounting to 39 . 0 percent of the total employed labour force. This initial imbalance increasesover the course of time: by 2009, women account for 48 . 5 percent of the labour force, yet theysupply merely 41 . 3 percent of the work volume. These figures strongly suggest that women whoentered the labour market in the post-1984 period worked relatively fewer hours than men, andpresumably also shorter hours than their female predecessors.
Proceeding to age groups, the entries show a moderately ageing workforce.[27] For both menand women, the fraction of workers under the age of 35 declines between 1984 and 2009, whereasthe proportion of workers older than 36 years of age increases or stays the same.[28] These trendsare somewhat more pronounced among women, supposedly because of their growing employmentshares which entail a ‘mechanical’ convergence between the demographic characteristics observedfor the workforce and the demographic structure of the population. This effect is potentiallyreinforced by a growing share of women returning into the labour market after the period ofchildbearing and child-rearing. One should notice that the decline of the relative share of youngworkers in general, and of the age cohort 21-25 in particular, exceeds the growth in overallemployment, and thus conforms to a decline in absolute figures. This effect is most significantamong men, but likewise clearly perceptible among women. Specifically, the data documentthat, between 1984 and 2009, the size of the age cohort 21-25 shrinks by 0 . 41 million men (from 1 . 24 to 0 . 83), and by 0 . 38 million women (from 1 . 44 to 1 . 06).
[...]
[1] Leaning on the kernel density reweighting method advanced by DiNardo, Fortin, and Lemieux (1996), the observed joint distribution of working hours may be conceptualised as a weighted average of group-specific hours distributions, where the weight of each group is determined by its relative size (cf. DiNardo et al. (1996), and Fortin, Lemieux, and Firpo (2011)). Changes in the workforce composition, that is, changes in observable group sizes, thus render the shape of the aggregate hours distribution by altering the relative weight which is attached to each group. Concomitant changes in the group-specific hours distributions affect the shape of the aggregate hours density by contributing different distributional features.
[2] Concerning the increase of high-skilled workers, Worldbank statistics document a rise of employees withtertiary degrees from 19.9 percent in 1992 to 27.5 percent in 2011 (Database accessed on May 21st, 2013). Thedata at hand reveal very similar estimates. With regards to the immigration waves, statistics provided by theFederal Ministry of the Interior (2011) and OECD.StatExtracts, accessed on May 15th, 2013, report a sharp rise in immigration figures and asylum applicants between 1991 and 1993. This inflow was initially dominated by refugees from Bosnia-Herzegovina and, later on, from Croatia.
[3] Eurostat Labour Force Statistics for the year 2008 (Database accessed on May 16th, 2013) report that 61 . 4 percent of German couple families are dual earners.
[4] For empirical support, see, for example, OECD (1999), Wolf (2000, 2010), Schmid (2010), and Sandor (2011). I will revisit the pros and cons of part-time work in due course.
[5] Cf. Schmid (2010), and Messenger (2007).
[6] For the sake of brevity, I leave a more comprehensive introduction of the GSOEP data to the next section, and confine myself at this point to what is necessary to motivate the issue.
[7] Even more abstract, income is the result of multiplying a certain amount of working time by a corresponding monetary remuneration per unit of working time.
[8] If the latter qualification holds true, observing daily wages is sufficient as it represents trends in both wages and incomes correctly. An explication of that matter will follow in due course.
[9] The data on average annual working hours for 13 OECD countries were drawn from the OECD.StatExtracts database on May 19th, 2013. They are described in the Appendix, and displayed comprehensively in AppendixTable A.1.
[10] Notably, a decline in annual working hours may, in principle, proceed from either an actual reduction ofhours per day or week, or from an expansion of vacation days, resulting in fewer days or weeks spent on regularwork.
[11] Notice that the illustrated periods are each 7 years apart. I decided to pool the data over four-year intervalsto facilitate a better representation of the hours distribution for a certain period. In an alternative specification,I used the following two-year intervals: 1984-1985, 1996-1997, and 2008-2009. The results did not change in avisually perceptible way.
[12] Notably, another famous figure in this context is the ‘28.8-hour week’ introduced at Volkswagen on January1st, 1994. One may view this as a success on the part of unions, yet it was achieved in the wake of a restructuringprocess at Volkswagen over the course of which an excess of approximately 30,000 workers was identified. Theeventual working time concessions of around 20 percent (1 day per week) roughly paralleled the annual incomecuts which workers had to incur. On the upside, however, the jobs were comprehensively secured. Source: http://labournet.de/branchen/auto/vw/stoeter1.html
[13] To forestall confusion, let me clarify in advance that the variable for overtime contains weekly figures, whilstthe survey question, as we will see later, requests the monthly overtime. The GSOEP uses a conversion factor of 4.3 weeks per month which I will adjust to 4.33 weeks per month for the purpose of my analysis.
[14] Notably, the 18th International Conference of Labour Statisticians held in 2008 proposed actual working hours as the key construct to estimate working time. In their definition of actual working hours, they explicitly promote the inclusion of overtime (cf. International Conference of Labour Statisticians (2008)).
[15] Working time accounts are understood as a means to flexibly displace working hours within predefined periods of time. Overtime, which is merely accumulated on such accounts and recompensed by time off does, by definition, not entail an expansion of individual working hours. Thus, if each and every worker gave an accurate reponse for the requested time period, the growing use of these tools would, in principle, not pose a threat to an accurate measurement of hours (cf. Zapf (2012)).
[16] The issue of travelling to and from work is more controversial than a first glance suggests. Essentially, the International Conference of Labour Statisticians (2008) states in its official guidelines that commuting is, in general, not to be seen as part of the working time. However, if the means of transport used for commuting permits - and is used - to work on the way, then it may count as working time.
[17] The SIAB data is certainly more renowned for the provision of highly accurate data on wages rather thanovertime hours. Nonetheless, the reporting remains in the hand of employers, and thus sidesteps, among otherthings, an imprecise memory of workers or intentional misreporting caused by a social desirability bias (cf. Boundet al. (2001)).
[18] The exclusion of workers that report to work less than 5 hours per week, that is, 1 hour per day, appearsreasonable and is affirmed by a study of Buddelmeyer, Mourre, and Ward-Warmedinger (2005) who conduct acomparative analysis of trends in part-time work for EU countries. The upper bound of 85 hours is somewhat ad hoc, and primarily serves to eliminate improbably high working hours that impair the ensuing computation ofwages from income data. Notably, the GSOEP data censor reportings above 80 hours per week. Values beyondthis threshold are thus rare, yet they occur in the current analysis due to the summation of standard hours and overtime.
[19] I researched several documents for an explanation for why the values in 1987 were not computed in quite the same fashion as in 1984 and 1985, but my efforts remained in vain.
[20] The corresponding GSOEP variable is called LABGRO$$. This variable benefits from a careful imputationof missing observations, and is consistently measured in EUR. For a detailed description, especially concerningthe imputation method, I refer to the documentation of generated variables provided by the DIW Berlin (2012).
[21] Bonus payments in the GSOEP are classified into six groups: 13th Month Pay, 14th Month Pay, Christmas Bonus, Holiday Bonus, Other (unspecified), and Profit Sharing Bonus. They are conjointly reported in retrospect which, firstly, necessitates a DM-EUR conversion for values before 2001, and secondly, requires to retrieve data for 2010, a year that is otherwise excluded from the analysis. Aggregate bonus payments are obtained from the sum of all individual bonus payments. Missing values are treated as zeros.
[22] In fact, if the degree of overstatement of working hours rises over the years, then wages would be increasinglydownward biased. This can impede an analysis of changes in wage inequality when trends in the degree ofoverstatement differ between different parts of the hours distribution. To give an example: if all workers overstatedtheir working hours by 10 percent in the first period, and by 20 percent in the second period, then, while thedownward bias in wages would increase between period one and two, relative wages between different workerswould stay the same, and thus changes in inequality that take effect between the two periods would also beestimated correctly. By contrast, if high income workers would overstate their hours by 20 percent in the secondperiod, but low income workers would maintain the 10 percent overstatement, then relative wages between highand low income workers change due to distinct trends in the degree of overstatement, and changes in inequalitybetween period one and two would be mismeasured. Notably, a comparison to macrodata shows a generally risingdegree of overstatement in GSOEP reportings on working hours (cf. Fuchs-Schündeln, Krueger, and Sommer(2010)). Yet, whether and how the degree of overstatement has evolved differently throughout the distribution ofhours remains an issue for further research.
[23] Individuals who report to be ‘in school’ are dropped.
[24] This presumes, justifiably, as I suspect, that the disutility of work partly descends from an inflexible timeframe within which work has to be carried out.
[25]Hence, the precise intervals are: 1984-1985, 1988-1990, 1993-1995, 1998-2000, 2003-2005, and 2008-2009.
[26] Initially surprised by the results, I found a good source of comparison in the OECD.StatExtracts database.The figures for the gender shares in civilian employment are virtually identical until 1999. In 2004 and 2009,the OECD statistics differ slightly, showing a share of female employment of 45 . 3 and 45 . 9 percent, respectively.However, for 2010, Wanger (2011) reports a female workforce representation of 49 . 8 percent for unified Germanybased on SIAB data. Noticing that female labour market participation is traditionally higher in East Germany(as is the share of full-time employed women), the figure lines up well with the documented results in Table 1.
[27] One should notice that subsuming ages in five-year intervals inevitably conceals changes in the distributionof ages within each age group. For example, suppose that, in 1984, the age group 36-40 contains five individualsaged 36, 37, 38, 39, and 40. The average age of this group is 38. In 1989, the composition of ages in this age groupchanges to 38, 39, 39, 39, 40. Accordingly, the average age is now 39. For the purpose of the present analysis, Ipropose to abstract from this imprecision, but for the sake of completeness, the possible occurrence of this errorshould be borne in mind.
[28] Men between 46 and 50 years of age pose a minor exception. The decline amounts to 0 . 2 percent.